web-dev-tools — router
The web-development toolkit is split three ways. Pick the discipline, then pick the skill inside it.
| Skill | Covers | Catalog |
|---|---|---|
| web-dev-frontend | Design systems & tokens, components, animation & motion, accessibility & UX review, SEO & Core Web Vitals, content & copy, frontend testing | 105 skills |
| web-dev-backend | API design & frameworks, databases & migrations, auth & access control, cloud & DevOps infrastructure, backend testing | 103 skills |
| web-dev-security | Web app & API pentesting, auth & session attacks, injection & request attacks, SAST & supply chain, API hardening, secrets, AI/LLM security | 180 skills |
Each file is self-contained — its own catalog, its own MCP table, its own rules. Loading only one keeps context small.
Step 1 — Which discipline?
| The request is mainly about… | Load |
|---|---|
| Visual design, layout, components, styling, brand, colour, typography | web-dev-frontend |
| Animation, scroll effects, transitions, motion, video | web-dev-frontend |
| Contrast, screen readers, keyboard nav, WCAG, UX review | web-dev-frontend |
| SEO, Core Web Vitals, Lighthouse, page weight, deploy | web-dev-frontend |
| Copy, docs, blog posts, decks, spreadsheets | web-dev-frontend |
| REST/GraphQL endpoints, framework choice, error handling | web-dev-backend |
| Schema, queries, migrations, indexes, cache | web-dev-backend |
| Login, sessions, tokens, OAuth, SSO, RBAC, IAM | web-dev-backend |
| Docker, Kubernetes, CI/CD, TLS, firewall, WAF, zero-trust | web-dev-backend |
| Backend tests, API integration tests, load & latency | web-dev-backend |
| Pentest or audit a web app / API | web-dev-security |
| Auth attacks — JWT, OAuth, BOLA/IDOR, mass assignment | web-dev-security |
| Injection classes — SQLi, XSS, SSRF, XXE, CSRF, prototype pollution | web-dev-security |
| SAST, dependency audit, SBOM, CI/CD supply chain | web-dev-security |
| Rate limiting, schema validation, gateway, API hardening | web-dev-security |
| Vault, secrets, key rotation | web-dev-security |
| Prompt injection, LLM guardrails, agent tool poisoning | web-dev-security |
Mixed task ("build a dashboard with auth"): load frontend + backend, but still pick only 1–3 skills total — one hub skill per discipline.
Ambiguous ("audit this app"): it is probably both an implementation review and a
security review. Load web-dev-backend for the code read and web-dev-security for the
finding. Say which skills you picked.
Step 2 — Which skill?
Read the discipline's SKILL.md catalog. Each entry is - \skill-name` — one-line description`.
Match the task, then ask the user to confirm with the question tool — never pick silently.
Offer the top 2–4 matches plus a "show the full catalog" escape.
Rule: load 1–3 skills, never the whole catalog. Hub skill first, then its sub-skills.
Step 3 — Which MCP?
| Task | MCP | Discipline |
|---|---|---|
| DOM, computed CSS, console, network, screenshots | chrome-devtools (call list_pages first for a pageId) | front + security |
| Accessibility score, Core Web Vitals, SEO audit | lighthouse | front |
| Flows, auth, multi-breakpoint capture | playwright | all |
| Current library/framework docs | context7 | front + back |
| Icons | lucide | front |
| shadcn/ui components | shadcn-registry | front |
| Generate images / video / audio | fal | front |
| Find design references | exa | front |
| Figma file → code | figma-official | front |
| Next.js | next-devtools | front |
| Diagrams / wireframes | excalidraw | front |
| Stock photos | images (needs API key) | front |
| Scrape pages & docs | firecrawl | all |
| Research current APIs / writeups | perplexity | all |
| CVE / EPSS / KEV | cve-intel | back + security |
| ATT&CK mapping | mitre-attack | security |
| Static analysis | semgrep | back + security |
| Template vuln scan | nuclei (needs NUCLEI_PATH) | security |
| Content / param fuzzing | ffuf (needs FFUF_PATH) | security |
| SQL injection | sqlmap (needs SQLMAP_PATH) | security |
| Host probing | httpx-scan (needs HTTPX_PATH) | security |
| Passive site intel | urlscan | security |
If an MCP call fails, fall back to the CLI equivalent and say so.
Step 4 — The loop
- Scope — target, stack, what "done" means.
- Inspect before you change — capture the current state (DOM, styles, screenshot, API shape) before editing. Never redesign or refactor blind.
- Design — tokens (front) or schema (back) before code.
- Build — real states, not placeholders. Match existing project conventions.
- Verify —
lighthouse+chrome-devtools+playwrightfor the front; tests +semgrep/codeqlfor the back;nuclei/ffuf/sqlmap+ verification skills for security. - Report — what changed, what was tested, what was not, residual risk.
Non-negotiables
- No visual claim without a before/after screenshot.
- WCAG AA contrast, verified with
lighthouse. - Real assets, no emoji as UI icons. Zero console errors when done.
- Respect
prefers-reduced-motion. - Security work: confirm authorisation scope first (bug bounty, CTF, own infra, localhost).
- Verify every finding before reporting it.
Commands
/web— pick any web-dev skill (asks which)/ui— pick a frontend skill (asks which)/backend— pick a backend skill (asks which)/sec— pick a security skill (asks which)/rea— pick from the full ~950-skill cybersecurity library (asks which)/cracked— pick an interview / career skill (asks which)/skills— search every installed skill